CAISI Will Test Google, Microsoft, and xAI Models Before Public Release
What Happened
The Center for AI Standards and Innovation (CAISI) signed agreements with Google DeepMind, Microsoft, and xAI for pre-release model evaluation. The deals follow earlier arrangements with OpenAI and Anthropic, meaning every U.S. frontier lab now offers the government a look before public launch. CAISI's evaluations focus on dangerous capabilities and national security risk.
My Take
This is regulation by procurement, not by statute, and it works because the government doesn't need a law — it just needs the labs to want federal contracts. The practical effect: a soft moat around U.S. frontier players. Anyone outside the CAISI tent (Chinese labs, scrappy open-source projects) gets locked out of federal procurement no matter how good their model is. If you're building on top of frontier models, the CAISI-evaluated list is going to look a lot like the FedRAMP list within two years. Plan your stack to match.
Read Original Source